extrait 0.3.1 → 0.4.0

package/README.md

@@ -233,6 +233,7 @@ Run examples with: `bun run dev <example-name>`
 Available examples:
 - `streaming` - Real LLM streaming + snapshot self-check ([streaming.ts](examples/streaming.ts))
 - `streaming-with-tools` - Real text streaming with MCP tools + self-check ([streaming-with-tools.ts](examples/streaming-with-tools.ts))
+- `abort-signal` - Start a generation then cancel quickly with `AbortSignal` ([abort-signal.ts](examples/abort-signal.ts))
 - `simple` - Basic structured output with streaming ([simple.ts](examples/simple.ts))
 - `sentiment-analysis` - Enum validation, strict mode ([sentiment-analysis.ts](examples/sentiment-analysis.ts))
 - `data-extraction` - Complex nested schemas, self-healing ([data-extraction.ts](examples/data-extraction.ts))
@@ -243,6 +244,7 @@ Pass arguments after the example name:
 ```bash
 bun run dev streaming
 bun run dev streaming-with-tools
+bun run dev abort-signal 120 "JSON cancellation demo"
 bun run dev simple "Bun.js runtime"
 bun run dev sentiment-analysis "I love this product."
 bun run dev multi-step-reasoning "Why is the sky blue?"

package/dist/index.cjs

@@ -1324,7 +1324,12 @@ async function executeMCPToolCalls(calls, toolset, context) {
       out.push({ call: metadata2, execution });
       continue;
     }
-    const args = isRecord(parsedArguments) ? parsedArguments : {};
+    const rawArgs = isRecord(parsedArguments) ? parsedArguments : {};
+    const args = context.request.transformToolArguments ? await context.request.transformToolArguments(rawArgs, {
+      name: toolName,
+      remoteName: tool.remoteName,
+      clientId: tool.clientId
+    }) : rawArgs;
     const metadata = {
       id: callId,
       type: call.type ?? "function",
@@ -1643,7 +1648,8 @@ function createOpenAICompatibleAdapter(options) {
           temperature: request.temperature,
           max_tokens: request.maxTokens,
           stream: true
-        }))
+        })),
+        signal: request.signal
       });
       if (!response.ok) {
         const message = await response.text();
@@ -1714,7 +1720,8 @@ async function completeWithChatCompletionsPassThrough(options, fetcher, path, re
       temperature: request.temperature,
       max_tokens: request.maxTokens,
       stream: false
-    }))
+    })),
+    signal: request.signal
   });
   if (!response.ok) {
     const message = await response.text();
@@ -1758,7 +1765,8 @@ async function completeWithChatCompletionsWithMCP(options, fetcher, path, reques
         tools: transportTools,
         tool_choice: request.toolChoice,
         parallel_tool_calls: request.parallelToolCalls
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -1823,7 +1831,8 @@ async function completeWithResponsesAPIPassThrough(options, fetcher, path, reque
       previous_response_id: pickString(body?.previous_response_id),
       temperature: request.temperature,
       max_output_tokens: request.maxTokens
-    }))
+    })),
+    signal: request.signal
   });
   if (!response.ok) {
     const message = await response.text();
@@ -1865,7 +1874,8 @@ async function completeWithResponsesAPIWithMCP(options, fetcher, path, request)
         tools: transportTools,
         tool_choice: request.toolChoice,
         parallel_tool_calls: request.parallelToolCalls
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -1941,7 +1951,8 @@ async function streamWithChatCompletionsWithMCP(options, fetcher, path, request,
         tool_choice: request.toolChoice,
         parallel_tool_calls: request.parallelToolCalls,
         stream: true
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -2043,7 +2054,8 @@ async function streamWithResponsesAPIPassThrough(options, fetcher, path, request
       temperature: request.temperature,
       max_output_tokens: request.maxTokens,
       stream: true
-    }))
+    })),
+    signal: request.signal
   });
   if (!response.ok) {
     const message = await response.text();
@@ -2125,7 +2137,8 @@ async function streamWithResponsesAPIWithMCP(options, fetcher, path, request, ca
         tool_choice: request.toolChoice,
         parallel_tool_calls: request.parallelToolCalls,
         stream: true
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -2689,7 +2702,8 @@ function createAnthropicCompatibleAdapter(options) {
           temperature: request.temperature,
           max_tokens: resolveMaxTokens(request.maxTokens, options.defaultMaxTokens),
           stream: true
-        }))
+        })),
+        signal: request.signal
       });
       if (!response.ok) {
         const message = await response.text();
@@ -2747,7 +2761,8 @@ async function completePassThrough(options, fetcher, path, request) {
       temperature: request.temperature,
       max_tokens: resolveMaxTokens(request.maxTokens, options.defaultMaxTokens),
       stream: false
-    }))
+    })),
+    signal: request.signal
   });
   if (!response.ok) {
     const message = await response.text();
@@ -2792,7 +2807,8 @@ async function completeWithMCPToolLoop(options, fetcher, path, request) {
         tools,
         tool_choice: toAnthropicToolChoice(request.toolChoice),
         stream: false
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -2875,7 +2891,8 @@ async function streamWithMCPToolLoop(options, fetcher, path, request, callbacks)
         tools,
         tool_choice: toAnthropicToolChoice(request.toolChoice),
         stream: true
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -4252,7 +4269,8 @@ async function callModel(adapter, options) {
     maxToolRounds: options.request?.maxToolRounds,
     onToolExecution: options.request?.onToolExecution,
     toolDebug: options.request?.toolDebug,
-    body: options.request?.body
+    body: options.request?.body,
+    signal: options.request?.signal
   };
   emitDebugRequest(options.debug, {
     provider: adapter.provider,

package/dist/index.d.ts

@@ -12,4 +12,4 @@ export { createOpenAICompatibleAdapter, type OpenAICompatibleAdapterOptions, } f
 export { createAnthropicCompatibleAdapter, DEFAULT_ANTHROPIC_MAX_TOKENS, DEFAULT_ANTHROPIC_VERSION, type AnthropicCompatibleAdapterOptions, } from "./providers/anthropic-compatible";
 export { DEFAULT_MAX_TOOL_ROUNDS } from "./providers/mcp-runtime";
 export { createDefaultProviderRegistry, createModelAdapter, createProviderRegistry, registerBuiltinProviders, type BuiltinProviderKind, type ModelAdapterConfig, type ProviderFactory, type ProviderRegistry, type ProviderTransportConfig, } from "./providers/registry";
-export type { CandidateDiagnostics, ExtractJsonCandidatesOptions, ExtractionCandidate, ExtractionHeuristicsOptions, ExtractionParseHint, HTTPHeaders, LLMAdapter, LLMRequest, LLMResponse, LLMStreamCallbacks, LLMStreamChunk, LLMToolCall, LLMToolDebugOptions, LLMToolExecution, LLMToolOutputTransformer, LLMToolChoice, MCPCallToolParams, MCPListToolsResult, MCPToolClient, MCPToolDescriptor, MCPToolSchema, LLMUsage, MarkdownCodeBlock, MarkdownCodeOptions, ParseLLMOutputOptions, ParseLLMOutputResult, ParseTraceEvent, PipelineError, StructuredAttempt, StructuredCallOptions, StructuredDebugOptions, StructuredError, StructuredMode, StructuredOptions, StructuredPromptBuilder, StructuredPromptContext, StructuredPromptPayload, StructuredPromptResolver, StructuredPromptValue, StructuredResult, StructuredStreamData, StructuredStreamEvent, StructuredStreamInput, StructuredStreamOptions, StructuredSelfHealInput, ThinkDiagnostics, ThinkBlock, StructuredTraceEvent, } from "./types";
+export type { CandidateDiagnostics, ExtractJsonCandidatesOptions, ExtractionCandidate, ExtractionHeuristicsOptions, ExtractionParseHint, HTTPHeaders, LLMAdapter, LLMRequest, LLMResponse, LLMStreamCallbacks, LLMStreamChunk, LLMToolCall, LLMToolDebugOptions, LLMToolExecution, LLMToolOutputTransformer, LLMToolArgumentsTransformer, LLMToolChoice, MCPCallToolParams, MCPListToolsResult, MCPToolClient, MCPToolDescriptor, MCPToolSchema, LLMUsage, MarkdownCodeBlock, MarkdownCodeOptions, ParseLLMOutputOptions, ParseLLMOutputResult, ParseTraceEvent, PipelineError, StructuredAttempt, StructuredCallOptions, StructuredDebugOptions, StructuredError, StructuredMode, StructuredOptions, StructuredPromptBuilder, StructuredPromptContext, StructuredPromptPayload, StructuredPromptResolver, StructuredPromptValue, StructuredResult, StructuredStreamData, StructuredStreamEvent, StructuredStreamInput, StructuredStreamOptions, StructuredSelfHealInput, ThinkDiagnostics, ThinkBlock, StructuredTraceEvent, } from "./types";

package/dist/index.js

@@ -1245,7 +1245,12 @@ async function executeMCPToolCalls(calls, toolset, context) {
       out.push({ call: metadata2, execution });
       continue;
     }
-    const args = isRecord(parsedArguments) ? parsedArguments : {};
+    const rawArgs = isRecord(parsedArguments) ? parsedArguments : {};
+    const args = context.request.transformToolArguments ? await context.request.transformToolArguments(rawArgs, {
+      name: toolName,
+      remoteName: tool.remoteName,
+      clientId: tool.clientId
+    }) : rawArgs;
     const metadata = {
       id: callId,
       type: call.type ?? "function",
@@ -1564,7 +1569,8 @@ function createOpenAICompatibleAdapter(options) {
           temperature: request.temperature,
           max_tokens: request.maxTokens,
           stream: true
-        }))
+        })),
+        signal: request.signal
       });
       if (!response.ok) {
         const message = await response.text();
@@ -1635,7 +1641,8 @@ async function completeWithChatCompletionsPassThrough(options, fetcher, path, re
       temperature: request.temperature,
       max_tokens: request.maxTokens,
       stream: false
-    }))
+    })),
+    signal: request.signal
   });
   if (!response.ok) {
     const message = await response.text();
@@ -1679,7 +1686,8 @@ async function completeWithChatCompletionsWithMCP(options, fetcher, path, reques
         tools: transportTools,
         tool_choice: request.toolChoice,
         parallel_tool_calls: request.parallelToolCalls
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -1744,7 +1752,8 @@ async function completeWithResponsesAPIPassThrough(options, fetcher, path, reque
       previous_response_id: pickString(body?.previous_response_id),
       temperature: request.temperature,
       max_output_tokens: request.maxTokens
-    }))
+    })),
+    signal: request.signal
   });
   if (!response.ok) {
     const message = await response.text();
@@ -1786,7 +1795,8 @@ async function completeWithResponsesAPIWithMCP(options, fetcher, path, request)
         tools: transportTools,
         tool_choice: request.toolChoice,
         parallel_tool_calls: request.parallelToolCalls
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -1862,7 +1872,8 @@ async function streamWithChatCompletionsWithMCP(options, fetcher, path, request,
         tool_choice: request.toolChoice,
         parallel_tool_calls: request.parallelToolCalls,
         stream: true
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -1964,7 +1975,8 @@ async function streamWithResponsesAPIPassThrough(options, fetcher, path, request
       temperature: request.temperature,
       max_output_tokens: request.maxTokens,
       stream: true
-    }))
+    })),
+    signal: request.signal
   });
   if (!response.ok) {
     const message = await response.text();
@@ -2046,7 +2058,8 @@ async function streamWithResponsesAPIWithMCP(options, fetcher, path, request, ca
         tool_choice: request.toolChoice,
         parallel_tool_calls: request.parallelToolCalls,
         stream: true
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -2610,7 +2623,8 @@ function createAnthropicCompatibleAdapter(options) {
           temperature: request.temperature,
           max_tokens: resolveMaxTokens(request.maxTokens, options.defaultMaxTokens),
           stream: true
-        }))
+        })),
+        signal: request.signal
       });
       if (!response.ok) {
         const message = await response.text();
@@ -2668,7 +2682,8 @@ async function completePassThrough(options, fetcher, path, request) {
       temperature: request.temperature,
       max_tokens: resolveMaxTokens(request.maxTokens, options.defaultMaxTokens),
       stream: false
-    }))
+    })),
+    signal: request.signal
   });
   if (!response.ok) {
     const message = await response.text();
@@ -2713,7 +2728,8 @@ async function completeWithMCPToolLoop(options, fetcher, path, request) {
         tools,
         tool_choice: toAnthropicToolChoice(request.toolChoice),
         stream: false
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -2796,7 +2812,8 @@ async function streamWithMCPToolLoop(options, fetcher, path, request, callbacks)
         tools,
         tool_choice: toAnthropicToolChoice(request.toolChoice),
         stream: true
-      }))
+      })),
+      signal: request.signal
     });
     if (!response.ok) {
       const message = await response.text();
@@ -4173,7 +4190,8 @@ async function callModel(adapter, options) {
     maxToolRounds: options.request?.maxToolRounds,
     onToolExecution: options.request?.onToolExecution,
     toolDebug: options.request?.toolDebug,
-    body: options.request?.body
+    body: options.request?.body,
+    signal: options.request?.signal
   };
   emitDebugRequest(options.debug, {
     provider: adapter.provider,

package/dist/types.d.ts

@@ -130,9 +130,11 @@ export interface LLMRequest {
     maxToolRounds?: number;
     onToolExecution?: (execution: LLMToolExecution) => void;
     transformToolOutput?: LLMToolOutputTransformer;
+    transformToolArguments?: LLMToolArgumentsTransformer;
     unknownToolError?: (toolName: string) => string;
     toolDebug?: boolean | LLMToolDebugOptions;
     body?: Record<string, unknown>;
+    signal?: AbortSignal;
 }
 export interface LLMUsage {
     inputTokens?: number;
@@ -192,6 +194,11 @@ export interface LLMToolExecution {
     durationMs?: number;
 }
 export type LLMToolOutputTransformer = (output: unknown, execution: Omit<LLMToolExecution, "output" | "durationMs">) => unknown | Promise<unknown>;
+export type LLMToolArgumentsTransformer = (args: Record<string, unknown>, context: {
+    name: string;
+    remoteName: string;
+    clientId: string;
+}) => Record<string, unknown> | Promise<Record<string, unknown>>;
 export interface LLMToolDebugOptions {
     enabled?: boolean;
     logger?: (line: string) => void;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "extrait",
-  "version": "0.3.1",
+  "version": "0.4.0",
   "license": "MIT",
   "repository": {
     "type": "git",
@@ -36,10 +36,10 @@
     "prepublishOnly": "bun run lint && bun run build && bun run build:types",
     "test": "bun test tests/ --reporter=dots --only-failures",
     "typecheck": "bunx tsc --noEmit",
-    "pack": "bun run build && npm pack"
+    "pack": "bun run build:types && bun run build && npm pack"
   },
   "dependencies": {
-    "@modelcontextprotocol/sdk": "^1.26.0",
+    "@modelcontextprotocol/sdk": "^1.27.1",
     "jsonrepair": "^3.13.2",
     "zod": "^3.25.76"
   },